Grant Holmes Welcome and good afternoon. I think we just made it over the top there. We are so pleased to have you all here today. My name is Grant and I'm your host for this webinar, the Road to Successfully Adopting AI. And we'll be talking today with subject matter experts in AI, specifically how it's being adopted. Please note that a few times during our meeting we'll drop in a few optional polls, but if you're so inclined to take a moment and answer them, we would really appreciate that. Just clicking a few buttons for us and you may submit a question at any time in comments and we will do our best to answer them during the session and or immediately afterwards or during the week via email. We appreciate all of the comments and questions. Well, today we're talking with leaders from KPMG and Microsoft on the road to successfully adopting AI and the impact AI is having on business and society advancements in AI like Microsoft Copilot are moving fast. So let's look at some of the common questions that come up. How does Copilot leverage existing data to train and learn in order to provide users with more powerful insights or recommendations, answers and solutions? And how can this technology bring next generation AI capabilities to every industry and business function? What are some key considerations when preparing for a successful implementation? So we're going to get to all that and more, but we want to drop in our first poll right away. So if you would, how did you hear about our webinar? Was it through Executive Leaders Network, KPMG, Microsoft Social Media? We really appreciate your answers and we'll leave that up just for a few moments. I'd like to have thank our panelists for joining in, and gents, if you would take a moment and introduce yourselves and then I'm going to toss in your first question.
Rob Moeller Sure, thank you. Oh, go ahead Satish.
Satish Paul Yeah, I can go first. Hi everybody. Thank you so much for attending this session. Satish Paul. I'm part of the global team here at Microsoft and I focus on professional services customers like KPMG, Accenture, Bain, McKinsey, et cetera, and also focus on architecture, construction and engineering customers. So today very happy to share my views and please if you need any follow-up sessions, very happy to engage in that too. Thank you.
Rob Moeller Hello everyone. I'm Rob Mueller and I'm a leader within our Microsoft practice. I focus on helping clients really leverage their Microsoft investment to achieve their business strategy and transform the way that they work, and that's been focusing on front, mid and back office, so I'm really looking forward to the conversation.
Grant Holmes Interesting. Well gentlemen, we are hearing a lot about innovative new technologies these days, including Microsoft Copilot. So to set the stage for this talk today, can you explain what this means?
Satish Paul Yeah, I can take a crack at that first. Thanks. So unless we have been living under a rock pretty much we have all heard about Chat GPT, where if you want to create a song and you want a few lyrics, you just type inside Chat GBT and it gives you the lyrics for the song. Or if you're trying to write a children's book, you can actually type the kind of story you want and Chat GBT prints it out. So what Microsoft did was it took the underlying technology called this OpenAI and made it enterprise grade and made it available inside our application. So that's what we call as Microsoft co-pilot. So over millennial human beings have been working side by side with humans, but now with Copilot would be able to have a computer or an assistant or a Copilot sitting next to us to help us with our day-to-day tasks. Some examples would be, if you look at the analyst’s data, what Gartner predicts is by 2020 there was about 1% of data generating generated using Generative AI. That is what the term is for technologies like OpenAI or Chat GBT, and by 2025 it's going to be more than 10% and the productivity boost just by turning on Copilot is in the range again from analysts at around 4 to 14%. And if you are a knowledge worker, you can focus on the human aspects of the task instead of focusing your time on mundane activities.
And the last thing I'll mention is Copilot is not one product, it's a term that is used and it's surfaced inside pretty much all the products from Microsoft. In my demo I'm going to show you how it works inside Word inside Excel. If you have a security product from Microsoft, it's available there and also if you have your own intellectual property, you can also surface that as a plugin so that you can monetize the IP that you have. So that's how I would define it.
Grant Holmes Fascinating, Rob?
Rob Moeller Yeah, I mean to add, as Satish mentioned, it is existing within the Microsoft applications and what this means to users is it's going to be a familiar interface, there isn't going to be as much learning as is way of helping them become more efficient and productive. It's not a tool that's going to replace workforces. Instead it's going to strengthen how they operate and what they can do. So it's built on the open AI Chat, GBT LLM, the large language model along with Microsoft Graph to enable these technologies.
Grant Holmes Rob, what is the large language? What does that mean, please?
Rob Moeller The large language model is the basis for AI. So in other words, you're going to go out and train a model. Microsoft is bringing that already. So the investment is, it's less for a client to be able to take advantage of it because it's coming built into their application. It's going to learn as they go as opposed to having to build their own models.
Grant Holmes Okay, thank you.
Rob Moeller They can converse and talk to the application and in that informal manner to generate answers and get the information they need more efficiently.
Grant Holmes Very interesting. Thank you for that. So I think it's time for us to get to a second poll if I'm correct. And we'd like to know where you are currently in your Copilot journey. You're not using it, you're using it, but you need more time or more info you're ready to purchase. Tell us a little bit more in that poll if you would. And again, we will leave that up for just a few minutes. So gentlemen, it sounds like this technology is presenting some exciting opportunities. So how should companies be preparing to use Copilot? Rob, you want to start off on this one?
Rob Moeller Sure. In 2023 we did KPMG did a Gen AI survey and right now it looks at about 78% of respondents believe that Gen AI will help them drive innovation. 65% believe it will help them gain a competitive advantage and 62% believe that it provides an opportunity for them to grow revenue share. So there's a lot of excitement and momentum now at the same time under 50% believe they are prepared to really get there with the tech talent and governance that they have in place. So there's a lot of preparation that they have to do. The way that we look at doing it is the first step is really review what you have from the Microsoft standpoint, what is your existing implementation? Understand how ready you are by looking at your strategy, your existing architecture guidelines and then weigh risk versus opportunity. Pick some use cases, start to identify some areas where you could truly test out within your Microsoft co-pilot instance where you'd gain value and look at those from the taking the existing models because that's what Copilot brings to start to bring some of those to life. That is going to be a less expensive way than going out and building new models and running from there and test every area that you go start to create really an AI team that can help to assess the benefits and what you're gaining from there along with minimize and mitigating your risks. You're going to have to truly look at what that means then as you put these use cases in from an OCM standpoint. Now because this is native, it's going to work within the constraints of Microsoft, but responsible AI is a framework that we have for controls, processes and tools that really helps you do and build out your AI journey in a trustworthy and ethical way. And that framework can help you set on the right path and the right starting the journey in the right way. Satish, do you have comments you'd like to add?
Satish Paul Yeah, yeah, thank you. That was great Rob. I'll add a couple of more things. The first thing is what I've seen work is employees enterprise wide have to be trained on how to interact with Generative AI because the better the questions, the better the answers and there are certain do's and don'ts. I'll give you one example. So working with one customer who interviewed lots of candidates, they gave offer letter, they were going to give offer letter for the successful candidates and the HR team felt that, well let's try to make it little bit of interesting the offer letter, let's make it interesting. And what they did was they took that offer letter for each of the candidates and copied and pasted that into Chat GBT and asked it to write that in a funny way and it did. So they took that and pasted that into the offer letter. They didn't know any better, but behind the scenes what happened is now Chat, GPT has all the knowledge about the people that they copied and pasted the offer details, how much salary, where they're living, all that detail information, detailed information is now part of the public knowledge as part of Chat GPT. So there are certain things you shouldn't do. And with Microsoft what we do is we don't use directly the public Chat GBT, we take an instance of that and instantiate that inside your Microsoft environment. So the data that you put inside OpenAI, which we call as Azure OpenAI is only available for you. It's not going to train the base model. So the first thing I would say is training is very important and the thing that comes after that is like Rob mentioned, turn on Copilot because that's the fastest way to value. There's not a lot of heavy lifting that you have to do, turn a few knobs and you can turn on Copilot. The next thing is you look at your existing applications and see where you can plug in generative AI. And the last piece is if you have some new applications that you're looking to build, then you can reimagine those applications using generating AI. So that's what I would say.
Rob Moeller I do like you adding the comment about the private data aspect. One of the big things is it's great with the Microsoft and the way that it's structured is every client asks me who owns the data and who owns this from here? Every client gets to keep complete control of their data, the security, and it does definitely reduce the risk.
Satish Paul Yep. Thank you
Grant Holmes Gentlemen. One of the things that came to my mind as you were talking is we've got micro businesses all over the world and then we've got huge corporations like Microsoft, et cetera. Who does this serve? Is there a good model or best model?
Rob Moeller I think every company, regardless of industry or size should be starting to evaluate and looking at how they can take advantage of AI. And again, it goes back to starting small and identify areas that are repetitive that are simpler to implement. Copilot as a part of Microsoft is being built into the product, they're going to be able to take advantage of this. They just a smaller organization may take a slower step, but at the same time a larger organization may put more governance in place and run from there. It's a balance, but any company is going to need to be conscious of this or aware in order to be successful and move towards the future.
Grant Holmes So in our first poll, the number of people that are not using this yet is fairly high. How would you encourage them to go about enacting this? I mean, do I go to a certain menu in Word and click on it or how is this enacted or Satish, will you get to that?
Satish Paul Yes, I will be showing that in my demo. A couple of places. Like I mentioned in the first question, Copilot is available in pretty much all the products and once it's turned on from the backend, the end users don't have to do a lot. You would see a button for Copilot to press that and it becomes available. So it’s very easy. Yep.
Grant Holmes I'm glad you're going to get to that. And we had an audience question about e-commerce and how it relates.
Satish Paul Is there any more detail on the question?
Grant Holmes Just if we're going to be talking about e-commerce, so I would assume some kind of integration there.
Satish Paul Yeah, I mean I can cover that Right now. The trend that we are seeing is autonomous agents. What that means is the autonomous agent would know my interests and then if I have to go and shop for something on Amazon, I don't actually go to amazon.com and buy the item. The autonomous agent does that on my behalf. It does all the typical things. It searches for the product that I'm interested in, looks at the reviews and all the things that a human would do the autonomous agent would go and do. So that's how we are seeing the industry move towards e-commerce. It doesn't matter what the e-commerce is at the backend, it is these autonomous agents which interact with those agents and gets the job done. So the human being doesn't have to keep doing the same thing because it not only can search, it looks at the result and analyzes the result pretty much like how a human would do. That's what I would say.
Rob Moeller And I can also see, I agree with that and I think from the e-commerce call center customer center running from there that the Chat bot and customer sentiment, you can start to assist and support a customer from purchasing from your organization in a more automated manner as well as gauging automatically their path, their interaction, their feelings on interacting with you so that when an issue or an item request does get to a human for remediation, they have context sensitive information and at the same time they can also use Copilot to find the answer across disparate data sources. So for example, if I'm in a manufacturing client and people want to come with a question about a part, I can use Copilot to search all of the information that I have available within my organization to automatically provide that information back with the most up-to-date information and help that customer be able to purchase from you.
Grant Holmes Okay. My burning question is, is there a plugin that when I go to Amazon it will actually show the product I actually search for?
Satish Paul I mean not at this time it is coming. That's where the development is happening, where if you are in the browser or any kind of interface, it's like an assistant for you. So yeah, it would know your interest and you just tell it what you want and it'll go through the steps that you normally go through.
Grant Holmes Great. Well let's move on with our questions here. Can you provide us with some examples of companies that are using Copilot to increase business value? Satish?
Satish Paul Yeah, yeah, thank you. Yeah, I can certainly go. And Copilot is not just for internal productivity scenarios, though, that is a big part of what Copilot can help you with. It also can be revenue generating. So we have customers who provide payroll services, generating payroll for the employees and there's a new service that they provided wherein the employee can upload their pay stub and ask questions just like they would ask a CPA, but they could ask this Gen AI interface for example, the employee says, well, I'm based out of Delaware, my business is I'm located in Delaware, but I want to move into Florida. What is the tax impact? So those kinds of questions they would be able to ask and get the answer. So that's a revenue generating service that the company was able to roll out. And a couple of other examples, if you have knowledge inside your organization, the earlier way of getting to that information is you search and you get a million results and it's your job to go through those million results and figure out where in those million results is the answer that you're looking for. But now using Gen AI very quickly you would be able to point to the corpus of information and then get to that one answer looking at all the documents. So that's also a very, very low hanging fruit. And finally, I think I'll cover that in my demo where if you have an RFP which is coming in that's bane of our existence, many times you have RFPs coming in, you reply to that and Gen AI could actually help you answer the RFPs a lot faster. It looks at the previous documents and drafts the rough draft of the output document or even if it's a proposal that you're trying to create. So those are some of the examples that I've seen.
Rob Moeller I see it a lot more with the customers that I talk to around knowledge management as sat talking about being able to finally get to all of the data sources without having to go through 17 screens and trying to find things on their portal and other data sources. So that is one of the biggest quick hits that I've seen. It's bringing that disconnected data together From a call center standpoint, it's a really good example of use cases you have, as I talked about, chat bot being able to provide automated responses and reduce and get to first call resolution from a lead management sales marketing standpoint, not having to dig through each lead and instead looking for letting AI figure out which ones are going to be the best ones for a seller to focus their time so they're not digging through things and instead they're talking to potential customers. As I also mentioned, customer sentiment, taking the interactions, engaging their mood, whether it's voice, whether it's email and tailoring responses and identifying ones that may need escalation or a tighter handholding from a marketing standpoint before you ended up creating a, there was very intensive to create marketing messages. Now you can take and define the criteria and ask Copilot to come up with potential emails and you can start to revise things based on different criteria. So the marketers are really focusing on refining and finalizing to get messages out rather than the initial content. It really is a limitless set of options. We look at anything where we're taking the data and trying to either do predictive analytics, refine things or write content. And this again is on all of the data that you own. And I did see there was a question in about insurance in the Copilot and looking at low market penetration with adopting AI, some of the areas from an insurance and financial services standpoint, looking at going back to your customer service, how you're interacting with them, looking at algorithms for fraud and risk management. Those are some other areas to start doing some insights on as well as optimizing your portfolio.
Satish Paul Yeah, just want to add a couple of points, Rob, you mentioned customer service. I can give the Microsoft example. We are use our own customer service capability and all we had to do was just turn on Copilot inside that and we have around 700,000 tickets open on every day, on every single day. So the productivity boost was more than 20% just by turning on Copilot and the backend team, the customer support team, they're optimizing the processes further because they're trying to shoot for 40%. So that's the kind of savings that you see in the customer service space and in the insurance, that's the market which is getting disrupted in a big way. For example, using Gen AI, insurance companies are able to target the high-risk customers in a very, very specific way, and now they're also able to say that, okay, you don't need insurance if your car is sitting in the garage. Only when you drive it gets triggered and you just pay for that period that you're using the car as an example. So it's going, it is getting disrupted in a big way. Just to give a couple of examples,
Rob Moeller You also made me think of that with the e-commerce and the question that came up before you start to look at the supply chain and where it is and being able to predict that and look at disruptions that are going to happen in your supply chain as well as being able to use that to adjust pricing. So I do like all of the areas. Again, if Satish and I sat here for the next hour, we could probably come up with 300 more use cases.
Satish Paul I agree. That's why we have a follow-up session. Right? Rob, you're going to talk about the MTC session where we can do a deeper dive, a half-day session, so this is just scratching the surface,
Rob Moeller Correct? Yeah, so as Satish is mentioning, what we're willing to do and offering is if anyone would like to talk and do a deeper workshop and the possibility of AMTC, which is the Microsoft Technology Center session, we would like to do that to help assess where you are and how you can get there on your journey.
Grant Holmes Fascinating, and what's the best, are we going to do contact information on that, Rob? Toward the end?
Rob Moeller Yes, we have a team that will be working and coordinating that so that Satish and I will be able to sit down and talk with the clients who are interested.
Grant Holmes Great. Fascinating. We had one other question. Looks like Copilot is going to be publicly available for EA companies. Is there an availability date for other type of Microsoft customers like CSP?
Satish Paul For CSP? I don't have the answer right now, but I can check and get back to you for the ea. I think it's November 1st.
Grant Holmes Okay, and one more. Is there any co-pilot use cases to search relevant policies and procedures? Are they stored in the Microsoft environment, can ask questions and be able to show relevant policies, natural language, et cetera?
Rob Moeller Yeah,
Satish Paul Definitely. Yeah, go ahead Rob.
Rob Moeller I was going to say yes, that is a very perfect use case for it just to start going through all of your policies and answering the questions, but also looking for areas to improve and to change. So yes, you have all the capabilities with that and it doesn't have to necessarily be stored in the Microsoft environment for the question. It has to be something that is in an accessible data source that can be pointed at.
Grant Holmes Okay, and is this ever local on your own machine or is it always in the cloud?
Satish Paul Yeah, it is primarily running from the cloud because there's a lot of computation that happens when it comes to AI.
Grant Holmes Well, Satish, we've talked several times about your demo. I think it's time to put you on stage and let's see it roll.
Satish Paul Okay, yeah, we don't have a lot of time. I'll just quickly show you what we are talking about Copilot in two products. So, the first one is the sales capability where if you're a salesperson, how can Copilot help you? And the second one is very easy, everybody could get access to. That is the word processor. If you're in Microsoft Word and you're trying to do a task, how can you make use of Copilot? Those are the two examples that I was going to present. Let me start with the sales one. So here if you're a salesperson, you get quite a bit of emails, you attend calls like this and the call gets recorded. Sometimes you're not able to do it on the call, but you want to be able to get the nuggets out of that call or maybe you're in a call and the call is happening, but then you don't want to be busy taking down notes. You want the system to understand what's being spoken and also give out the action items. Those are the kind of things that sales Copilot provides. So if you're a salesperson that's provided out of the box, let me show you a quick example of that. If I'm in my email right now and I look at this email and I'm trying to reply to this, and you would see that there's quite a bit of information in here. There's an email chain going back and forth. I can go in here and on the right-hand side you see Copilot where it's giving me prompts, it says that, Hey, do you want to reply to this inquiry? Do you want to offer a discount? Do you want to make a proposal or do you have a custom prompt that you want to use to draft the email? You would be able to select any of those prompts. And then sales co-pilot behind the scenes, OpenAI Azure OpenAI is doing all the crunching. It's looking at this email thread and it's coming up with the answer. So it's not only looking at the email thread, it also goes to the backend system and it figures out, so this person who is contacting me wants to know the price about a particular product. It picks up the information from the backend system, and it might have taken maybe five minutes for me, but with this sales co-pilot experience, I can get that done in a few seconds. I select the appropriate prompt and if I want to tweak it, I can say, okay, can you rewrite it and it's going to rewrite that for me, and if I like this, I can say copy this, it copies that into the email, and then if I want, I can edit this and then I can just click on send. So this is the experience for the salesperson where when they get these kinds of emails and if they want to quickly draft an email, Copilot would be able to do that. Again, this is just one capability of what sales co-pilot can allow you to do, and then you also have the ability to do quite a bit of additional things. Let me go into this one. I'll go into my Word example here. So I'm in my Word document and I have Copilot available, so that's the Copilot.
Grant Holmes We're not currently seeing your Microsoft Word document. We lost the screen somehow.
Satish Paul Okay. Thank you for letting me know. Let me share now. Can you see my screen now? The word,
Grant Holmes Well, we've got a white screen.
Satish Paul Okay, just give me one second.
Rob Moeller No problem. If you'd like. Well, there was another question that came in while we were doing that. I'll go ahead and try to address that. It was around support for FDA and HIPAA regulations for pharmaceutical and healthcare use cases. If you're already working within Microsoft in the dynamic space and the other Azure environments, then you're going to end up following those security models. For example, if you're running Cloud for Healthcare, it's going to support it in that same manner. The data is still in your cloud and running from there. So those same regulations would apply, would be something that we could probably have a deepest conversation about because understanding your landscape and how it really is structured would be essential to validating that. Go ahead, Satish.
Satish Paul Yeah, can you see my Word?
Grant Holmes Yes, we can.
Satish Paul Okay. Yeah, so what I was showing was in my word file, I see the Copilot icon. I can click on that and when I'm inside this, the prompt comes up and the prompt I can say is, Hey, create an employment contract for a temporary hire based off New York law. If that's the contract that I'm trying to create, I can go to Copilot inside Word and I can just type it, type the prompt, and then there's always something, let me open a new document. So I pasted that prompt in there and then what you will see is it's going to make use of the Azure OpenAI, which is the new law and the associated contract information, which would be in a temporary hire situation, and it gives me this contract so I can take this as a,
Grant Holmes We're not seeing any results we're seeing. I think you opened up a new document and that's what we were seeing. Okay,
Satish Paul You should be seeing this contract now. Yep,
Grant Holmes There we go.
Satish Paul Okay. Yeah. So I just put that into the prompt and it immediately gave me this output and this could be used as a starting point, and then if I wanted to tweak co-pilot to tweak it, it'll be able to tweak it. So that's how easy it is. Copilot is just sitting there. Whenever you need help, you can just click on that and you would be able to leverage the capability that Copilot provides. So I just want to give you two examples and it's available in pretty much all the products from Microsoft and it really speeds things up for you. And many times you're trying to create a new application, you can totally reimagine the user experience, you reimagine the kind of application that you're trying to build, making use of capabilities like this
Grant Holmes A lot of time and a lot of grief right there. That's really wonderful. Did we want to go ahead and post our third poll in the background, gentlemen, if it didn't happen already? That third poll is would you be interested in joining a live Copilot workshop at a Microsoft training or technology center MTC in the future? And we would love to know of that and we will follow up over the next period short period of time and make sure everybody's involved. Is that correct, gentlemen?
Rob Moeller Yes. And it'll be on a client by client and customer by customer basis. We're not going to do this as a broader one. What we want to do is do a deeper dive into where you are in the journey and how to get started
Grant Holmes Very specific to that customer. Great. Okay, so I've got some other questions here. We've got plenty of time left. How do we get started? How does a company get started?
Satish Paul Do you want to take that? Yeah, please.
Rob Moeller This is my favorite part. It's that first workshop. Let's sit down and truly assess where you are, what you want to do to develop that roadmap and strategy. At that point, as we talked about picking some areas to start with, but also looking at your data. I mean, if your environment and your data is in a bad state, there may be different inputs and precursors that have to be done to start, but that is really architecture data where you are as an organization and what you can really be able to absorb. It's not something where you just turn it on and it runs. It's something where it's going to take care and feeding, and there needs to be a structured process in place. One of the nice things about KPMG is we're one of the first firms that did Gen AI and working on that journey, we've been doing it for years. As I said, we have a responsible AI framework and a very mature AI practice along with deep industry practices. So we bring all of that power to sit down and assess and understand where you are, where you want to get, and come up with a comprehensive plan to get you there.
Grant Holmes Okay, so this isn't something like thesaurus or the bell check, you just don't turn it on and start using it. It's going to require a little bit more.
Rob Moeller You could turn it on. I would recommend that cases I would really look at is, it goes back to the governance and the other aspects of it. It's a tool and you need to be able to have the proper training and understanding on how to use the tool.
Grant Holmes And we've got a question on the law example. You did Satish. Is the law in public domain or stored internally?
Satish Paul Yeah, it's pulling it up. It's pulling from the public domain and is the point I was mentioning in my first question, if you have custom ip, and that's what one of our customers, Thomson Reuters, they're big in the legal space. What they did was they took their IP and packaged that as a custom plugin. So the same way as I clicked on the Copilot, I would be able to leverage the Thomson Reuters and then it would go into all the knowledge repository information that is available from tr, and that would be used for crafting the contract. And that's a paid service, so you are able to monetize your IP. So it's not just the public domain, that's the one that I showed, but you can also package your information and you can tap into that. Then it'll be more specific domain expertise.
Grant Holmes And the first question I think of is, do I still have to have an attorney go over it after it does all that?
Satish Paul Yeah, I would say so. Yeah. That's why we don't call it autopilot, we call it Copilot, right? The human is always the final decision maker. We can speed things up, but then the human has to approve it. So absolutely.
Rob Moeller It's never a human replacement.
Satish Paul Got it.
Rob Moeller People need to be trained and still verify.
Grant Holmes So how do you start small? And to me it seems like you're going to have to have some size companies so you can have some people dedicated to the AI journey or team. How do you create a strategy? How do you grow it? Is there,
Rob Moeller It really comes back to sitting down and developing that strategy, and we talked about the simple use cases and running from there. It's start with understanding what you have, where you want to get to, and then pick some low risk areas. You do not want to run into some of the situations where you're trying to have it automate your legal answers and not have the oversight. You maybe want to respond to email queries and put in small chat bots and other isolated areas to start to measure time savings, accuracy and efficiency, and then adjust and configure based on what you find.
Grant Holmes Okay. So it sounds to me like I think I want to do X with AI, but after a conversation with you, I realize it'd be nice to step back one time or four times and get this going first, and that probably is the value of the meeting. Is that accurate?
Rob Moeller Yeah, I mean, yes it is. You're going to find how much time it takes to maintain and monitor and govern the pieces along with what policies you need to change and you're going to step in and iterate through that. So yes, I do think it needs to be it thoughtful approach as opposed to just jumping in and turning things on.
Grant Holmes Gotcha. Well, we want to remind everyone, I'm hoping we still have our poll up for those of you that might be interested in a live workshop. So that is still up, and if it's not, maybe we can turn it back on. We can keep answering that. It sounds like that's the thing to do. So let's talk about ethics, ethical ai, responsible practices and frameworks, a key PMG. Rob, you want to look at that?
Rob Moeller Sure. And as I mentioned, we have an entire responsible AI framework. At a high level, you look at eight areas that we cover in the whole practice around this. It's fairness, it's getting the data from free of bias. It is explainability being able to understand and explain the documentation and the algorithms, the accountability, driving the ownership and responsibility across all of the lifecycle security. That's a big one with everyone, is making sure that you're not getting unauthorized access or corruption to the data. Along with the privacy aspect, every organization is concerned with making sure their data's private and compliant. You have to then as well weigh in the safety, making sure there isn't a negative impact to humans, to property, to environment and et cetera. And all along that the data needs to maintain integrity, so you have to have trust with that data and it has to have reliability. So those are the top level areas and those are the tools to truly develop in a trustworthy and ethical manner. And that gets you to the ability to accelerate value and there's many more components to it that we would dig into deeper with and bring in the right people from that responsible AI lens.
Grant Holmes Okay. We, we've kind of touched, I mean Satish, did you want to add anything on to that? I'm sorry.
Satish Paul No, no. I think that is very important because if something is happening and something unethical is happening, then customers lose trust. For example, in Microsoft Azure, we have this capability where I can talk for five minutes and Azure would listen to how I'm speaking, and after that, whatever text I give, it'll speak just like how I'm speaking. So for that, Microsoft has an ethics officer assigned to each customer. Not everybody can go and avail their features. They make sure that it's being used the right way. For example, if you go to the insurance company, there's this famous personality, the lady, they recorded her voice and now when you call the call center, the answer is coming from that person. So ethics is very, very important. Otherwise it spoils the reputation of the company.
Grant Holmes Interesting stuff. Thank you. We talked a little bit about human oversight. Anything more that you would like to talk about there, gentlemen?
Rob Moeller Other than just to reinforce that it is required and mandatory, this is never going to be a replacement. This is always something that there needs to be intervention and validation. It is a tool to make you more productive.
Grant Holmes I think we've said that plenty of times. So that's good. We've got two questions that kind of go down the same line. Well, I will Microsoft, see the information I'm posting, do I have privacy? And then who owns the IP? I think those kind of go together.
Satish Paul Yeah, I'll take a crack and maybe Rob you can add to it, right? So when you are posting a question into our out of the box services that we provide, like the sales co-pilot that I provide, that could be considered as a SaaS offering or software as a service offering. It's prepackaged, it's readily available. The same thing for word example, the word example that I showed, it's prepackaged or you can create custom. That's a lot more work. So in that case, what happens is in both these scenarios, it goes into the Azure OpenAI and our engineers look at the questions which are coming in and it's saved for 30 days, but if you feel that that shouldn't happen, you would be able to request Microsoft and you can turn off. So none of the information which is coming in, and the reason why we are looking at is to make sure that hey, is the prompts coming in being answered the right way? It's not to mine and get any insights out of the prompts that you're asking, but even if you want to turn that off, you would be able to turn it off. So yeah, it's not using your data and the sales example that I showed the email, it's just taking the email extract and then going into the backend system, taking all of that, that is the prompt for the word example. I put in one prompt saying, Hey, go per new, create a contract, the same thing. It pulls up this whole email and sends that as the prompt to OpenAI, OpenAI processes it, Azure, OpenAI, and then gives the reply, and after that the data is deleted so it doesn't get persisted at all. So that's why it's very important to realize that Microsoft, in all the co-pilots that we have, we are making use of Azure OpenAI not directly going against the public OpenAI. Anything else you want to add?
Rob Moeller No, I think that was a really good summary.
Grant Holmes Thank you gentlemen. Will the multimodal LLMs be available soon, sort of like text images, audio, video, et cetera?
Satish Paul Yes, it is coming out. You might have seen it. If you're going to Bing Chat inside the Edge browser, you have the capability to have Bing chat where you can click on the icon and it turns on a Chat GPT like experience. And in that today, if you go in and type a text, and if the text, the reply requires some image, it actually calls the image creation. So the multimodal features are getting rolled out right now, the text along with images that are happening, but slowly the other modes would also be added.
Grant Holmes Fascinating. The company, the question came from we are starting simple with our AI approach and leveraging AI to deliver personalization without data experiences for on the B2B side, drive digital journeys based on size of company, industry location to increase engagement. I'm sure that's more of a statement than a question. Would that all be good stuff?
Satish Paul Yeah, yeah, absolutely. I mean, many of the products already has Copilot from Microsoft. For example, if it's a journey creation that you're focused on, then you can say that, Hey, I just had a meeting with a customer and this is the journey that I want to put them in. Create a journey for that customer. You just type the prompt and automatically the journey gets built and of course it's Copilot, so you can take it and quickly tweak it and then you can launch it. So after that, the meeting is over and maybe the first step is you send a summary of the meeting, all the steps would be defined in the journey, and AI could come in handy there because you just type a prompt and the whole journey gets created as an example. But there are other areas too where you would be able to generate, I mean, leverage and AI.
Grant Holmes So if I'm someone in a fairly large company controlling all of my servers and all this data, enterprise architecture, how is this going to impact that?
Satish Paul So we also have AI available in, like I mentioned earlier too, in pretty much all of our products. So if you're going into the security capability that we provide, we have a co-pilot in there. So if you want to monitor what's happening, the log files inside your applications, you would be able to leverage that co-pilot and ask for what's happening. And then it can also give you suggestions on how to fix it if there's a problem, or maybe one piece of it is getting hacked, it'll actually give you suggestions on how to fix it, and sometimes you can also go and fix that for you. So that is how I would leverage, so the Copilot is available in all facets,
Grant Holmes But it's not going to affect my architecture. I'm not going to have to rewire my data room or things like that. This all happens outside, is that?
Satish Paul Yeah, so looking at you would have to point to the data sources, and after that it's pretty seamless, like how I clicked on the Copilot button icon, you would be just clicking on that and the interface shows up and you'll be able to leverage it.
Grant Holmes Okay.
Rob Moeller I would add that I do think that you need to look at your enterprise architecture. You need to see how things are going to move through your environment along with any other AI and other changes that are coming. Again, you want to maintain data integrity, security, reliability, et cetera. It comes down to that responsible AI framework and how data's flowing through it is something not just to turn on and let loose. So I do want him just reinforce that.
Satish Paul Yeah.
Grant Holmes You just mentioned something, Rob, that I could see becoming a nightmare. One division or set of people over here decides to put on Copilot and another one decides to put on this one, and all a sudden you've got management issues of competing data. Is that a problem or has that not happened?
Rob Moeller It could be. It is something to be conscious of. I mean, you get into some of the other areas like hallucinations and data, AI taking inferring to AI and getting to different answers and looking at how the data moves. So it is something to be conscious of. It falls back to that entire look at the strategy of where you're trying to go and where you want to start.
Grant Holmes So yet another good reason to have a good meeting with you guys and figure out where we're at and where we're going. Sounds like,
Satish Paul Yeah, I mean, Gen AI is good at making stuff up, right? Hallucination drop mentions, so we have to be careful.
Grant Holmes Touch on that for a moment. That's a new concept I hadn't heard of.
Satish Paul Yeah, yeah. So what I'm referring to here is the AI has been trained with pretty much all the public corpus of information. And if you're asking a question and if Gen AI is not trained with that information, just like a human, it's going to infer what should be the answer for that question, and it comes up with an answer. And sometimes what we find is that answer is not the right answer. So it's hallucinates and makes stuff up. So that's why we have to be very careful, and I think that's the point that Rob was mentioning. So you have to have the guardrails put in place. There should be a property valuation if you're going to leverage it,
Grant Holmes Guardrails and blinders, great analogies, does it work with existing permissions within your business applications and stuff?
Rob Moeller For example, if you're putting it in, if Copilot is being turned on within Dynamics power platform or CE and you're running from there, you're going to inherit that same security model. Same way across the permissions with the office applications. It's still something that I would review and make sure that all of your security models are up to date within those tools because you're putting new functionality in place.
Grant Holmes I have a vague question here that I'm not sure I understand. COE for AI
Rob Moeller Center of excellence.
Grant Holmes Okay, help us with that.
Rob Moeller Yeah, so that's what we talked about earlier when you're getting started setting up that governance and that team and you could establish it as ACOE to be able to help drive the adoption and the progression of your AI journey.
Grant Holmes Sounds like a creation of a roadmap as it were, to a point.
Rob Moeller To some extent, yes.
Satish Paul Okay. Yeah, roadmap and also on a day-to-day basis, what are the guardrails that you have to put in place so that the usage is monitored, and if there's some usage which is not recommended, then there is some correction mechanism. So yeah, I think KP MG would be providing this kind of COE services to make sure that it's going the right way.
Grant Holmes Very good. Jerry wanted to know, are there any use cases regulators adopting AI in regulatory activities?
Rob Moeller There are, and I can dig into that and then follow up on that to Jerry because yes, there are definitely some use cases. I just don't have the context and familiarity with them firsthand right now, but I can put someone in touch with them.
Grant Holmes Very good. Sean wanted to know about documents on responsible AI architecture?
Rob Moeller Yes, we have those as well as a part of the responsible AI framework and running from there, and we'd be happy to sit down and talk in that more detail and bring in the people from that team.
Grant Holmes Great. Well, we want to encourage our audience. We have several minutes left here if you have any other questions, gentlemen, is there anything you'd like to loop back on or expand on that you think is important?
Rob Moeller I think Go ahead, Satish.
Satish Paul No, no, please go ahead.
Rob Moeller I was going to say I think it's been a good conversation and the questions have been great, and I think we've covered a broad area of topics, so I appreciate and thank everyone for attending.
Satish Paul I just want to add one piece. So you saw what's happening in the EV space, electric vehicles, their Tesla started more than 10 years ago, and now pretty much everybody is using their charging standard, and so they have the AI maximum miles already run, so their models are trained. That's going to be much better because most of these systems are based on reinforced learning. So if you are, let's say, doing tax filing and if you are able to train your model with more customer scenarios, so your model is going to be much better than your competition who started late, so you might want to really look at it early so that you're not left behind because after a point you just cannot catch up because the new entrance or the people who went in first, they were able to train their models and it's far advanced than what you have, so you might want to seriously look at it and see if you are interested in adopting it.
Grant Holmes Those are solid points. Satish, Cheryl came up with one. Any best practices to share for financial institutions on treading safely?
Rob Moeller I think a lot of the ones that we've talked about. First of all, the way that you start and looking at how you're going to govern it with the financial data and institutions, I'm working with one now, and that is always top of mind. It is a very process of stepping slowly through and not trying to do too much because of the aversion to risk. So it's just like a healthcare organization and the ones that I work with from that, it is a very methodical and well thought out process to identify an area, move into a test, adjust and then continue. So it's not a getting from zero to a hundred in a short time. It is a little longer of a journey.
Grant Holmes I have another question from the audience. Can users use Copilot on customer face as customer facing agents?
Satish Paul Yeah, they can. So if it's a chat bot that you have created, then you would be able to surface that and they would be able to interact with that, and you also have the ability for creating your own Copilot experience, and then you could be surfacing that inside your own applications.
Grant Holmes Yeah, fascinating.
Rob Moeller There was one more question that I don't think got addressed.
Grant Holmes Okay.
Rob Moeller It was the company, do companies own the IP rights to the generated materials? That is probably something to talk to with your legal representation and your legal department within an organization, because I don't think Satish or I are comfortable being attorneys and understanding where the sources are, but it is something that is beyond the technology and more of a legal decision.
Satish Paul I just want to add one point there, right from Microsoft side, because that is a big concern, right? There's a lot of data, public domain, which has been used to train the model, but now if you take that data and then you generate some new document or some new image or new content, are you going to get sued by somebody who's the original owner even though it was in the public domain? So Microsoft provides legal indemnity. That's something that is provided, so maybe you can speak with your Microsoft folks and they'll be able to share more details on that. So we try to protect our customers saying that, Hey, if you're using Azure OpenAI and something happens and somebody comes after you, Microsoft has it covered.
Grant Holmes Okay, great. Which industries are slow adopters and what are the roadblocks?
Rob Moeller One industry that sticks out in my mind is healthcare. It is something because of the risk and HIPAA is the privacy aspect. It is going to be a slower journey for them. Technology companies, I see the exact opposite. They're going to be the early adopters and running faster on it. It's the same way that I would compare it to when cloud migration started. The organizations that were slow to move to the cloud may be the ones that are slower, and those industries may be the ones that are slower to move towards AI, but with the amount of clients in the industries that I talk to, it isn't that in every space, clients are either wanting to get prepared or wanting to start.
Satish Paul I'll just add one point to that. What I see is then the industry is more manual in nature where the information or the IP is not captured in documents or systems, then they're slow to adopt because there's no purpose of information that they can point to. They're all going to get disrupted, but there's a little bit of slowness when it comes to those industries, is what I see. Great.
Grant Holmes Well, gentlemen, we're coming to a close on our time here. I'm going to give you one last chance for a wrap. Anything else to hit?
Rob Moeller No, but other than we'd be happy and look forward to sitting down and doing a workshop and an MTC session with people.
Grant Holmes Great, great. Well, we'd like to thank both of you, Rob Moeller from KPMG and Satish Paul from Microsoft for your expertise today, the executive leaders network team for arranging the event and most importantly to you, the delegates and guests who without your participation, these events would not be able to take place over the next few days. The KPMG team will reach out to you to invite you to continue this conversation as it's been offered many times, and address any other questions that may have not been answered. I can't imagine what, but probably very specific stuff, and that's what we'll do. Please feel free to take this opportunity for some free advice that is more personalized to your company's individual needs. Gentlemen, thank you for being here and to our attendees, thank you for attending and good luck and goodbye.
Satish Paul Thank you. Thank you everybody.
Some or all of the services described herein may not be permissible for KPMG audit clients and their affiliates or related entities.
The information contained herein is of a general nature and is not intended to address the circumstances of any particular individual or entity. Although we endeavor to provide accurate and timely information, there can be no guarantee that such information is accurate as of the date it is received or that it will continue to be accurate in the future. No one should act upon such information without appropriate professional advice after a thorough examination of the particular situation.
The KPMG name and logo are trademarks used under license by the independent member firms of the KPMG global organization.
@2023 KPMG LLP, a Delaware limited liability partnership and a member firm of the KPMG global organization of independent member firms affiliated with KPMG International Limited, a private English company limited by guarantee. All rights reserved.
Microsoft and Microsoft Azure are trademarks of the Microsoft group of companies.
